Online Program

Detection of Anomalous Activities in Social Networks Using Scan Statistics, Multivariate Time Series, and Topic Modeling (308523)

Keywords: Anomalous Activity, Scan Statistics, Topic Modeling, Multivariate Time Series

Excessive communications and dynamic relationships between influential individuals or nodes in a social network are important to understand the behavior and pattern of the communication networks. Here a multivariate time series model, vector autoregressive model has been developed and applied to the metadata of organization e-mails as a case study to detect a group of influential nodes and their dynamic relationship. The objective of the present work is to identify the dynamic relationship of an ego with other egos in the neighborhood ego networks. Particularly, we are interested in revealing who is initiating the excessive communication, when it is occurring, and how it is being diffused among other nodes in the neighborhood. Furthermore, we investigate what is being discussed excessively among nodes by utilizing both the probabilistic topic model, Latent Dirichlet Allocation, and scan statistic model. We demonstrate how the influential vertices obtained from the VAR model are connected with the anomalous topic activities. These analyses provide new ways of detecting the excessive communications and anomalous topic flow through influential nodes in a dynamic network.